Discussion Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)

[Vattila]

Except for the details about the improvements in the microarchitecture, we now know pretty well what to expect with Zen 3.

The leaked presentation by AMD Senior Manager Martin Hilgeman shows that EPYC 3 "Milan" will, as promised and expected, reuse the current platform (SP3), and the system architecture and packaging looks to be the same, with the same 9-die chiplet design and the same maximum core and thread-count (no SMT-4, contrary to rumour). The biggest change revealed so far is the enlargement of the compute complex from 4 cores to 8 cores, all sharing a larger L3 cache ("32+ MB", likely to double to 64 MB, I think).

Hilgeman's slides did also show that EPYC 4 "Genoa" is in the definition phase (or was at the time of the presentation in September, at least), and will come with a new platform (SP5), with new memory support (likely DDR5).

What else do you think we will see with Zen 4? PCI-Express 5 support? Increased core-count? 4-way SMT? New packaging (interposer, 2.5D, 3D)? Integrated memory on package (HBM)?

Vote in the poll and share your thoughts!

 

[gdansk]

Put on your tinfoil hat:
N5 (Zen 4) + N6(IO die) = 11
Double RDNA and Zen announcement = 11 * 2 = 22
22/11/11
it all makes sense, wake up people

I'm joking, I don't even know if they'll use 6nm on either

 

[DrMrLordX]

Paper launch!

Even that requires a launch-level UEFI absent bugs as well as a few post-QS CPUs for demos (followed by little actual stock). Not sure the OEMs are quite ready. I could be wrong.

I'm joking

Good thing, too. November would be a slog.

 

[ryanjagtap]

Put on your tinfoil hat:
N5 (Zen 4) + N6(IO die) = 11
Double RDNA and Zen announcement = 11 * 2 = 22
22/11/11
it all makes sense, wake up people

I'm joking, I don't even know if they'll use 6nm on either

Well, with zen 4 having igpu on its IO die, and RDNA2 products on N7 and N6, we can speculate that the IO die may be N7 or N6, so that they don't have to validate RDNA2 on other node.

 

[biostud]

Well, with zen 4 having igpu on its IO die, and RDNA2 products on N7 and N6, we can speculate that the IO die may be N7 or N6, so that they don't have to validate RDNA2 on other node.

Since it is all new design, there would hardly be any reason not to go N6.

 

[A///]

It will hurt Intel more if AMD attacks from the rear.

All red and rocketing forward.

 

[A///]

Good thing, too. November would be a slog.

November would be an absolute crap move. I'm thinking late summer or pushed up to July release, because 7th month 7000 series if AMD is still into that goofy stuff. Releasing a new set in November when they can't match the supply Intel can would only lead to the same thing that happened in 2020 with the release of Zen 3. Sold out for months. And you really don't want that to happen when people can get decent deals on hardware with the holiday season in full swing.

Seems like a no brainer to me, but this is also the company that underestimated the take rate of the 3900X and 3950X. They'd presumed, if I'm remembering Clark's interview correctly, that while they were impressive chips at the time, they didn't consider that most people wouldn't want to pay Intel rates.

 

[ryanjagtap]

Since it is all new design, there would hardly be any reason not to go N6.

That, and N6 is I think cheaper than N7 (maybe, not sure).

 

[Hans Gruber]

I don't understand the logic here. We are talking 5nm TSMC process. It's not a new fab. They could pump out millions of chips in a few months. AMD needs to release Zen 4 ASAP. A 7/7 launch at the latest is still late if you go by the original release schedule. Times change and you need to be able to adapt to the market.

 

[RnR_au]

I don't understand the logic here. We are talking 5nm TSMC process. It's not a new fab. They could pump out millions of chips in a few months. AMD needs to release Zen 4 ASAP. A 7/7 launch at the latest is still late if you go by the original release schedule. Times change and you need to be able to adapt to the market.

Why do they *need* to release Zen 4 ASAP? They are ahead in most of the markets. See Markfw's post earlier in the thread for an Intel-AMD summary.

 

[Hans Gruber]

Why do they *need* to release Zen 4 ASAP? They are ahead in most of the markets. See Markfw's post earlier in the thread for an Intel-AMD summary.

They don't need to release it early. But when Intel accelerates their new chips and changes their roadmap. Well, we can come back to this post at the end of the year. I don't want to say that I told you so. If AMD releases Zen 4 mid year. They would have the market lead in every category for several months. People forget Alder Lake is 10nm. Every other Intel CPU was 14nm before it. Soon Intel will be on 7nm. People forget the efficiencies Intel has gained with the current (10nm) die shrink. I think 7nm will be more significant. Add to that significant architecture enhancements of their CPU's.

I just think it's important that AMD has a strong early showing with AM5. An all new chipset with DDR5. If Intel pulls a move, they always have. People building all new systems will have a harder time going with AMD if they are in 2nd place. I guess it doesn't matter.

Look at RDNA2 vs. Nvidia 30 series cards. AMD thought they would be further along in their battle with Nvidia but the 30 series smokes the AMD cards. They say RDNA3 will be well ahead of Nvidia. Maybe, but like Intel. Nvidia adapts to market conditions. AMD makes much better GPU's than before RDNA but those were value play cards (RX-580).

On the server side, AMD has nothing to worry about because of how efficient their chips are compared to Intel.

 

[Mopetar]

I don't understand the logic here. We are talking 5nm TSMC process. It's not a new fab. They could pump out millions of chips in a few months. AMD needs to release Zen 4 ASAP. A 7/7 launch at the latest is still late if you go by the original release schedule. Times change and you need to be able to adapt to the market.

I'm not sure I understand your logic. It's not as though AMD is completely unable to compete at the moment. They're the top choice for server and really the only one for HEDT, and in desktop they're generally competitive with Intel until you get below the ~$200 mark.

I think it's far more important for AMD to take their time and making sure that the AM5 rollout goes off without hitch and that they've got plenty of supply built up so that we don't see a repeat of prior launches where there were board issues or too little supply. Intel isn't going to be releasing anything new before AMD's launch that might take away customers and waiting only helps improve the DDR5 situation.

 

[RnR_au]

Soon Intel will be on 7nm. People forget the efficiencies Intel has gained with the current (10nm) die shrink. I think 7nm will be more significant. Add to that significant architecture enhancements of their CPU's.

I don't track Intel closely, but I did come across this slide earlier in the year from Intel themselves...



"Up to double-digit performance boost"... I think you are somewhat optimistic in regards to the Intel 7 process and their arch enhancements

But as always wait for benchmarks!

 

[Hans Gruber]

I don't track Intel closely, but I did come across this slide earlier in the year from Intel themselves...

View attachment 60222

"Up to double-digit performance boost"... I think you are somewhat optimistic in regards to the Intel 7 process and their arch enhancements

But as always wait for benchmarks!

Intel 7 is still 10nm but advanced/enhanced 10nm. Intel 4 is 7nm.

 

[DrMrLordX]

I don't understand the logic here. We are talking 5nm TSMC process. It's not a new fab. They could pump out millions of chips in a few months. AMD needs to release Zen 4 ASAP. A 7/7 launch at the latest is still late if you go by the original release schedule. Times change and you need to be able to adapt to the market.

While they are late releasing Raphael, they aren't necessarily late sending Genoa to hyperscalars in limited quantity, which is their primary focus. Regardless, the rumour mill seems to agree that dice arrived at packaging facilities this month for Raphael, indicating an August launch.

 

[RnR_au]

Not sure how late is 'late'...

Zen 1 Mar 2017
Zen 1+ Apr 2018 - 13months - special case... minor tweaks
Zen 2 Jul 2019 - 15 months
Zen 3 Nov 2020 - 16 months
Zen 4 Aug 2022 - 21 months

Maybe a little late... but given the sheer amount of gold the cloud guys/gals are handing over to AMD, I can understand the server first focus.

 

[moinmoin]

It was mentioned before that Genoa is already shipping to the first customers, since early this or late last month I think?

 

[deasd]

Ok, this is what Arctic said: *by the end of 2022*

https://twitter.com/x/status/1516349654409134083

and this:

https://twitter.com/x/status/1516331139434090496

 

[gdansk]

I guess there's a downside of AMD being able to repurpose their CCDs to server customers who pay more and have nearly unlimited appetite. We get fewer and/or delayed desktop parts.

 

[CakeMonster]

Arctic is just saying '2022' in an awkward way, I think people are reading too much into it.

 

[Ajay]

Ugh! After 2 & 1/2 years of speculation, I just want to see Raphael in action. 4-5 month wait is a killer at this point.

 

[Saylick]

Ugh! After 2 & 1/2 years of speculation, I just want to see Raphael in action. 4-5 month wait is a killer at this point.

We'll see more in a little over a month.

 

[gdansk]

Ugh! After 2 & 1/2 years of speculation, I just want to see Raphael in action. 4-5 month wait is a killer at this point.

Psh, you think you're impatient? We really impatient people are already not waiting for Zen 5. Scrap Zen 4, we demand Zen 5 in 2022

 

[Mopetar]

I guess there's a downside of AMD being able to repurpose their CCDs to server customers who pay more and have nearly unlimited appetite. We get fewer and/or delayed desktop parts.

Not necessarily. The higher revenue also means they can buy more wafers from TSMC and invest more in R&D for better future products.

 

[Ajay]

Not necessarily. The higher revenue also means they can buy more wafers from TSMC and invest more in R&D for better future products.

Still, AMD is wafer limited at TSMC. TSMC just can't keep up with demand for it's top customers on leading edge nodes (not enough EUV supply from ASML and it's vendors, for example).

 

[Mopetar]

Having more money puts them in the best possible position to be able to purchase any additional capacity that becomes available. Or to potentially buy wafers from anyone else that TSMC is supplying.